WordNet based measures

[José Martínez]

Hello,

How can I run a WordNet-based semantic similarity measure ? 

I'm trying to compile dkpro with maven but maybe the dependencies are no longer available. Is there a compiled Jar ? could you please give me some recommendations and/or examples?

thank you,

José

[Torsten Zesch]

Dear José,

the WordNet-based measures require the DKPro LSR project that is currently not on maven central due to license issues.

However, you should be able to use it when either putting the LSR project in your workspace or installing the dependencies manually.

-Torsten

--
You received this message because you are subscribed to the Google Groups "DKPro Similarity Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dkpro-similarity-users+unsub...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[José Martínez]

Hi Torsten,

Thanks for your answer, I installed DKPro dependencies manually and was able to run the similarity comparator based on WordNet 3.0. My main concern now is whether the results are correct because I obtained slightly different results to the ones provided by http://maraca.d.umn.edu/cgi-bin/similarity/similarity.cgi (WordNet::Similarity WordNet 3.0). For example, given the following pair of words and running WuPalmerComparator (for my DKPro implementation and the Web service), my results are as follows: 

Words DKPro(local) WordNet::Similarity

car-automobile 1.0 1.0

gem-jewel 1.0 1.0

journey-voyage=0.8571428571428571 0.9565

boy-lad=0.7692307692307693 0.9524

coast-shore=0.7272727272727273 0.9231

asylum-madhouse=0.8571428571428571 0.9565

magician-wizard=1.0 1.0

midday-noon=1.0 1.0

furnace-stove=0.42105263157894735 0.5714

food-fruit=0.26666666666666666 0.4706 

Do you believe that such results are fine according to different algorithm implementations? is there a way that my local compilation went wrong and the results vary?

Please let me know your opinion or recommendations and sorry for the inconvenience. 

Best,

José

To unsubscribe from this group and stop receiving emails from it, send an email to dkpro-similarity-users+unsubscri...@googlegroups.com.

[Torsten Zesch]

Hi José,

interesting.

I don't think it is a problem with your version. DKPro Similarity returns those results and I have currently no idea how to interpret the variance.

Is there also a difference in the other measures, or is it only WuPalmer?

-Torsten

To unsubscribe from this group and stop receiving emails from it, send an email to dkpro-similarity-users+unsub...@googlegroups.com.

[José Martínez]

Hi Torsten, 

using Lin Comparator the results are:

Words DKPro(local) WordNet::Similarity

car-automobile=1.0 1.0

gem-jewel=1.0 1.0

journey-voyage=0.8403386817522845 0.8277

boy-lad=0.8504397740207656 0.7979

coast-shore=0.9852307446559548 0.9632

asylum-madhouse=0.9684051384601088 0.9813

magician-wizard=1.0 1.0

midday-noon=1.0 1.0

furnace-stove=0.21814634256802945 0.2294

food-fruit=0.16127400912797896 0.1559

Thanks for your comments.

[Torsten Zesch]

Hi José,

at least the values are a bit closer. 

There could always be subtle differences in the implementations. When we developed the library the measures, we did not work from a certain implementation, but from the description in the papers.

We might also use a slightly different WordNet version.

-Torsten

[Torsten Zesch]

Another comment: for debugging it would be interesting if you could share the resources.xml you are using for your experiments.

-Torsten